Nodes and Connections in the AI Semiconductor Stack
This report turns the v2 semiconductor AI alpha bundle into a practical network interface: 100 ranked public nodes, 2,422 relationship rows, 800 display edges, depth-five path maps, source trails, cash-flow caveats, and diligence actions for deciding which connection is signal and which is only ecosystem adjacency.
50 US and 50 non-US public companies across memory, foundry, packaging, tools, materials, ASIC, analog, and EDA.
Each row preserves relationship type, use case, cash-flow direction, evidence label, confidence, and source/target ranks.
Path chains expose how apparently distant public names connect back through foundry, OSAT, HBM, metrology, equipment, and materials layers.
Samsung Electronics carries the highest derived centrality score at 71.1.
The bottleneck thesis is intact - hyperscalers now guide $630B+ of 2026 capex (+62% y/y, ~75% AI-specific), Microsoft pins $25B of it on memory/component inflation, and DDR5 contract prices have more than doubled - but the alpha has rotated from discovered US connectivity into the memory-capex derivative chain. June 9's tape makes the point: SK hynix +15.9% and Samsung +9.0% in the Korea melt-up while Coherent fell 12.0% and Marvell 10.4% in the third US optics flush since May 7, even as Marvell raised FY27/28 outlooks. The cleanest mispricing is Hanmi at rank 3 with just +4.5% YTD the day after SK hynix handed it a 44.2bn-won HBM4 TC-bonder order, while the page still crowns Credo ($41B cap, +61.5% YTD) and Astera Labs ($54.3B, +79.8% YTD) - exactly the crowded names the ranking's own crowding penalty should be demoting. The substrate tier (Samsung Electro-Mechanics +18.4%, Nan Ya PCB +10.0%, Ibiden +9.0% today) is entirely absent from the list and is the obvious structural gap. The active ranking pass cross-checks the original thesis against current market prices, market cap, YTD rerating, AI 2027 compute/security assumptions, hyperscaler capex pressure, power constraints, HBM and advanced-packaging evidence, and physical-AI adoption signals.
Use the network to find where alpha and dependency disagree.
The useful action is not simply buying the highest-ranked ticker. The better workflow is to compare alpha rank, graph centrality, relationship confidence, cash-flow visibility, and market-cap crowding, then decide whether the node is a hidden tollgate, a crowded bottleneck, or a loose adjacency.
The ranking is not the same as the graph.
Credo and Astera lead the alpha screen because smaller market caps, high AI connectivity exposure, and stronger valuation torque can dominate megacap incumbency. Samsung, Broadcom, TSMC, GlobalFoundries, and Tower surface as central nodes because the network measures how much traffic passes through a company, not just how under-discovered it is.
Most edges are real but not invoice-level.
28 edges are tagged high-confidence and 1760 are medium-confidence. The page treats undisclosed cash amounts explicitly: a supplier relationship, equipment dependency, or capacity bottleneck is investable signal, but it is not the same as disclosed revenue by customer.
Centrality is a diligence queue, not a buy list.
High centrality tells you where to ask harder questions: customer concentration, wafer starts, CoWoS or OSAT allocation, HBM qualification, capex cadence, inventory risk, export controls, and whether a company can keep price while capacity normalizes.
Interface actions that make the map useful
The page is structured as a research terminal: first inspect the visual graph, then rank nodes, then audit relationship rows, then trace depth-five paths, then confirm the cited source. That sequence prevents a broad supply-chain story from collapsing into a generic semiconductor basket.
A node can be high alpha, high centrality, or both.
The map below emphasizes the top centrality nodes and the highest-quality display edges from the bundle. Hover or focus a node to compare rank, centrality, inbound and outbound edge load, and tier role.
Samsung Electronics
Memory rebound plus HBM4/SOCAMM2/NVIDIA Vera Rubin evidence and foundry optionality. Size and conglomerate complexity cap alpha despite improving AI linkage.
Centrality quick picks
Mindmap from demand to diligence
A visual compression of how the graph should be read.
AI Demand
Capacity Owners
Tooling and Control
Second-Order Scarcity
Diligence Decision
Rank, centrality, tier exposure, and confidence say different things.
The charts deliberately separate alpha score from network load. A high-alpha company can be a focused interconnect winner, while a centrality leader may be a large capacity owner with less valuation torque.
Top 20 alpha scores
The screen favors smaller AI-connectivity, test, packaging, and equipment names with stronger discovery torque.
Alpha vs mapped public connections
Size is represented by market cap. The upper-right quadrant is high-alpha plus high-connection density.
Relationship composition
Ecosystem adjacency, fab sales, foundry use, packaging/test, equipment, and materials dominate the edge set.
Confidence stack
The graph is intentionally conservative: confidence is visible before any row is used as a thesis input.
US vs non-US by tier
The bundle is global: non-US capacity, memory, packaging, equipment, and materials nodes remain central to the AI buildout.
Most common tier lanes
A lane is a source-tier to target-tier edge count. It shows which layer pairs explain the most mapped relationships.
Tier-to-tier lanes reveal which parts of the stack carry the most mapped dependency.
The layer flow table groups the full relationship graph by source tier and target tier. It is useful for spotting where relationship density comes from broad ecosystem mapping versus hard supply-chain choke points.
Top layer lanes
Named constraints to read first
HBM and memory
SK hynix, Micron, and Samsung are high-signal nodes because AI memory demand has direct platform relevance, visible pricing cycles, and qualification bottlenecks.
Foundry and packaging
TSMC, Samsung, GlobalFoundries, OSATs, and packaging specialists sit between AI chip demand and deliverable units. Watch capex, allocation, and customer mix.
Tools, test, and metrology
Equipment and inspection nodes convert capacity announcements into yields. These names often carry second-order upside before the market sees customer-specific revenue.
Interconnect and custom ASIC
Credo, Astera, Marvell, Broadcom, and peers connect AI clusters and custom accelerators. The ranking rewards focused exposure and valuation torque.
The ranking table keeps price, market cap, links, cash summaries, and alpha reason together.
Use the table to compare a company rank against its tier, region, public connection count, current price field from the bundle, and relationship-derived diligence notes.
| Rank | Node | Tier | Score | Cases | Price | Market cap | Links | Cash lens | Alpha reason |
|---|---|---|---|---|---|---|---|---|---|
| #1 | Hanmi Semiconductor 042700.KQ / KOSDAQ / Non-US | OSAT / packaging | 100.0 central #13 | 151,000 | $14.0B | 68 mapped 25 direct | In: Cash in from wafer sort, final test, packaging, burn-in, advanced packaging a... Out: Cash out to bonding/test/inspection equipment, substrates, labor, facilities ... | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #2 | Camtek CAMT / NASDAQ / Non-US | wafer fab equipment | 96.2 central #50 | 140 | $6.53B | 28 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #3 | Amkor Technology AMKR / NASDAQ / US | OSAT / packaging | 95.1 central #12 | 66.91 | $16.6B | 68 mapped 25 direct | In: Cash in from wafer sort, final test, packaging, burn-in, advanced packaging a... Out: Cash out to bonding/test/inspection equipment, substrates, labor, facilities ... | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #4 | Nova NVMI / NASDAQ / Non-US | wafer fab equipment | 91.8 central #47 | 452 | $14.4B | 28 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #5 | Credo Technology CRDO / NASDAQ / US | custom ASIC / interconnect | 91.7 central #4 | 259 | $48.2B | 79 mapped 32 direct | In: Cash in from hyperscaler custom programs, data-center networking/optical conn... Out: Cash out to foundries, OSATs, IP/EDA, substrates, inventory and R&D. | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #6 | BE Semiconductor Industries BESI.AS / Euronext Amsterdam / Non-US | OSAT / packaging | 90.4 central #14 | 251 | $24.7B | 68 mapped 25 direct | In: Cash in from wafer sort, final test, packaging, burn-in, advanced packaging a... Out: Cash out to bonding/test/inspection equipment, substrates, labor, facilities ... | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #7 | Alchip Technologies 3661.TW / TWSE / Non-US | custom ASIC / interconnect | 90.3 central #26 | 4,210 | $9.98B | 43 mapped 32 direct | In: Cash in from hyperscaler custom programs, data-center networking/optical conn... Out: Cash out to foundries, OSATs, IP/EDA, substrates, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #8 | Coherent COHR / NASDAQ / US | analog / power / RF | 90.2 central #65 | 317 | $62.0B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #9 | ASM International ASMI.AS / Euronext Amsterdam / Non-US | wafer fab equipment | 86.0 central #56 | 924 | $51.0B | 21 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #10 | Astera Labs ALAB / NASDAQ / US | custom ASIC / interconnect | 85.1 central #9 | 393 | $67.4B | 77 mapped 32 direct | In: Cash in from hyperscaler custom programs, data-center networking/optical conn... Out: Cash out to foundries, OSATs, IP/EDA, substrates, inventory and R&D. | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #11 | King Yuan Electronics 2449.TW / TWSE / Non-US | OSAT / packaging | 81.4 central #55 | 309 | $12.1B | 25 mapped 20 direct | In: Cash in from wafer sort, final test, packaging, burn-in, advanced packaging a... Out: Cash out to bonding/test/inspection equipment, substrates, labor, facilities ... | public semiconductor test-services provider added because AMD filings identify KYEC as an outsourced assembly/test partner; AI upside depends on utilization, test intensity and advanced packaging pull-through. | |
| #12 | Onto Innovation ONTO / NASDAQ / US | wafer fab equipment | 81.3 central #36 | 291 | $14.5B | 39 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #13 | FormFactor FORM / NASDAQ / US | wafer fab equipment | 80.8 central #35 | 112 | $8.70B | 39 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #14 | Global Unichip 3443.TW / TWSE / Non-US | custom ASIC / interconnect | 79.4 central #27 | 4,340 | $18.7B | 43 mapped 32 direct | In: Cash in from hyperscaler custom programs, data-center networking/optical conn... Out: Cash out to foundries, OSATs, IP/EDA, substrates, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #15 | Advantest 6857.T / TSE / Non-US | wafer fab equipment | 78.3 central #37 | 29,160 | $131.9B | 39 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #16 | Tower Semiconductor TSEM / NASDAQ / Non-US | foundry | 77.9 central #7 | 217 | $26.1B | 78 mapped 24 direct | In: Cash in from wafer starts, long-term capacity agreements, advanced-node wafer... Out: Cash out to equipment/material suppliers, fab construction, utilities, labor ... | manufacturing tollgate in AI supply, with score influenced by pricing power, technology node, and geopolitical risk. | |
| #17 | Navitas Semiconductor NVTS / NASDAQ / US | analog / power / RF | 77.3 central #66 | 13.36 | $3.25B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #18 | Entegris ENTG / NASDAQ / US | materials / wafers | 76.4 central #20 | 139 | $20.7B | 61 mapped 20 direct | In: Cash in from recurring consumables and wafer/material volumes tied to fab uti... Out: Cash out to raw materials, energy, specialty chemicals, capex and process eng... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #19 | TOWA 6266.T / TSE / Non-US | OSAT / packaging | 75.9 central #45 | 4,710 | $438M | 36 mapped 25 direct | In: Cash in from wafer sort, final test, packaging, burn-in, advanced packaging a... Out: Cash out to bonding/test/inspection equipment, substrates, labor, facilities ... | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #20 | ASPEED Technology 5274.TW / TWSE / Non-US | analog / power / RF | 75.5 central #63 | 13,665 | $18.6B | 22 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #21 | Disco 6146.T / TSE / Non-US | OSAT / packaging | 75.4 central #51 | 71,510 | $48.1B | 25 mapped 25 direct | In: Cash in from wafer sort, final test, packaging, burn-in, advanced packaging a... Out: Cash out to bonding/test/inspection equipment, substrates, labor, facilities ... | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #22 | ACM Research ACMR / NASDAQ / US | wafer fab equipment | 74.1 central #57 | 98.07 | $6.80B | 21 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #23 | Lasertec 6920.T / TSE / Non-US | wafer fab equipment | 74.0 central #70 | 43,020 | $26.5B | 20 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #24 | Eugene Technology 084370.KQ / KOSDAQ / Non-US | wafer fab equipment | 73.6 central #71 | 147,000 | $2.31B | 20 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #25 | GlobalFoundries GFS / NASDAQ / US | foundry | 73.4 central #6 | 67.97 | $37.3B | 78 mapped 24 direct | In: Cash in from wafer starts, long-term capacity agreements, advanced-node wafer... Out: Cash out to equipment/material suppliers, fab construction, utilities, labor ... | manufacturing tollgate in AI supply, with score influenced by pricing power, technology node, and geopolitical risk. | |
| #26 | Shibaura Mechatronics 6590.T / TSE / Non-US | OSAT / packaging | 73.3 central #52 | 5,300 | $2.43B | 25 mapped 25 direct | In: Cash in from wafer sort, final test, packaging, burn-in, advanced packaging a... Out: Cash out to bonding/test/inspection equipment, substrates, labor, facilities ... | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #27 | Qualcomm QCOM / NASDAQ / US | analog / power / RF | 73.0 central #53 | 187 | $196.6B | 25 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #28 | Socionext 6526.T / TSE / Non-US | custom ASIC / interconnect | 72.6 central #29 | 2,664 | $3.12B | 43 mapped 32 direct | In: Cash in from hyperscaler custom programs, data-center networking/optical conn... Out: Cash out to foundries, OSATs, IP/EDA, substrates, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #29 | Broadcom AVGO / NASDAQ / US | custom ASIC / interconnect | 72.2 central #2 | 389 | $1.85T | 82 mapped 34 direct | In: Cash in from hyperscaler custom programs, data-center networking/optical conn... Out: Cash out to foundries, OSATs, IP/EDA, substrates, inventory and R&D. | Custom AI accelerators, Ethernet networking and infrastructure software cash flow. Strong platform, but high market cap, customer concentration and visibility reduce early-alpha score. | |
| #30 | Marvell Technology MRVL / NASDAQ / US | custom ASIC / interconnect | 71.4 central #5 | 232 | $260.8B | 79 mapped 32 direct | In: Cash in from hyperscaler custom programs, data-center networking/optical conn... Out: Cash out to foundries, OSATs, IP/EDA, substrates, inventory and R&D. | picks-and-shovels exposure to AI connectivity, HBM/advanced packaging, metrology, or ASIC ramps with smaller market visibility than the megacaps. | |
| #31 | NVIDIA NVDA / NASDAQ / US | AI accelerator | 71.0 central #8 | 204 | $4.94T | 79 mapped 27 direct | In: Cash in from hyperscalers, cloud providers, OEM/ODM server channels and enter... Out: Cash out to foundries, memory suppliers, OSATs, substrates, ODMs, inventory c... | AI cash-flow king with disclosed dependencies on TSMC/Samsung foundries and SK hynix/Micron/Samsung memory. Alpha is capped by megacap crowding, ETF ownership and very visible expectations. | |
| #32 | Siltronic WAF.DE / XETRA / Non-US | materials / wafers | 70.7 central #59 | 88 | $3.17B | 21 mapped 20 direct | In: Cash in from recurring consumables and wafer/material volumes tied to fab uti... Out: Cash out to raw materials, energy, specialty chemicals, capex and process eng... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #33 | Taiwan Semiconductor Manufacturing TSM / NYSE ADR / Non-US | foundry | 70.3 central #3 | 437 | $2.48T | 78 mapped 24 direct | In: Cash in from wafer starts, long-term capacity agreements, advanced-node wafer... Out: Cash out to equipment/material suppliers, fab construction, utilities, labor ... | core AI winner, but already highly owned; score is capped by crowding and valuation discovery. | |
| #34 | Tokyo Ohka Kogyo 4186.T / TSE / Non-US | materials / wafers | 70.3 central #60 | 10,125 | $8.07B | 21 mapped 20 direct | In: Cash in from recurring consumables and wafer/material volumes tied to fab uti... Out: Cash out to raw materials, energy, specialty chemicals, capex and process eng... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #35 | Parade Technologies 4966.TW / TWSE / Non-US | analog / power / RF | 69.7 central #72 | 655 | $1.55B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #36 | NXP Semiconductors NXPI / NASDAQ / Non-US | analog / power / RF | 69.7 central #73 | 284 | $71.7B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #37 | Semtech SMTC / NASDAQ / US | analog / power / RF | 69.2 central #77 | 129 | $12.0B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #38 | Tongfu Microelectronics 002156.SZ / SZSE / Non-US | OSAT / packaging | 69.2 central #64 | 72.17 | $15.2B | 20 mapped 20 direct | In: Cash in from wafer sort, final test, packaging, burn-in, advanced packaging a... Out: Cash out to bonding/test/inspection equipment, substrates, labor, facilities ... | public outsourced semiconductor assembly/test provider added because AMD filings identify the ATMP joint ventures with Tongfu; upside linked to AMD compute demand and packaging/test capacity. | |
| #39 | Shin-Etsu Chemical 4063.T / TSE / Non-US | materials / wafers | 68.3 central #48 | 7,010 | $80.8B | 28 mapped 20 direct | In: Cash in from recurring consumables and wafer/material volumes tied to fab uti... Out: Cash out to raw materials, energy, specialty chemicals, capex and process eng... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #40 | MKS Instruments MKSI / NASDAQ / US | wafer fab equipment | 68.1 central #69 | 349 | $23.6B | 20 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #41 | Infineon Technologies IFX.DE / XETRA / Non-US | analog / power / RF | 67.8 central #75 | 72.32 | $118.3B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #42 | Micron Technology MU / NASDAQ / US | memory | 65.8 central #15 | 949 | $1.07T | 52 mapped 34 direct | In: Cash in from HBM/server DRAM/eSSD contracts, spot/contract pricing and OEM/cl... Out: Cash out to equipment makers, materials/wafer suppliers, fab construction, EU... | HBM and data-center memory cash flow with public revenue/margin acceleration and contracted HBM supply visibility; less crowded than GPU leaders but increasingly discovered. | |
| #43 | Teradyne TER / NASDAQ / US | wafer fab equipment | 65.4 central #38 | 352 | $55.0B | 39 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #44 | SK hynix 000660.KS / KRX / Non-US | memory | 65.2 central #16 | 2,186,000 | $1.33T | 51 mapped 34 direct | In: Cash in from HBM/server DRAM/eSSD contracts, spot/contract pricing and OEM/cl... Out: Cash out to equipment makers, materials/wafer suppliers, fab construction, EU... | HBM/server DRAM/eSSD leader with record profits and AI-led pricing power. Strong alpha, though recent share-price records mean the crowd is catching on. | |
| #45 | MACOM Technology Solutions MTSI / NASDAQ / US | analog / power / RF | 65.0 central #78 | 305 | $23.3B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #46 | Samsung Electronics 005930.KS / KRX / Non-US | memory | 63.4 central #1 | 278,000 | $1.40T | 87 mapped 35 direct | In: Cash in from HBM/server DRAM/eSSD contracts, spot/contract pricing and OEM/cl... Out: Cash out to equipment makers, materials/wafer suppliers, fab construction, EU... | Memory rebound plus HBM4/SOCAMM2/NVIDIA Vera Rubin evidence and foundry optionality. Size and conglomerate complexity cap alpha despite improving AI linkage. | |
| #47 | Monolithic Power Systems MPWR / NASDAQ / US | analog / power / RF | 63.3 central #79 | 1,316 | $67.9B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #48 | Alpha and Omega Semiconductor AOSL / NASDAQ / US | analog / power / RF | 62.8 central #82 | 37.37 | $1.30B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #49 | Novatek Microelectronics 3034.TW / TWSE / Non-US | analog / power / RF | 62.7 central #81 | 542 | $10.0B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #50 | Tokyo Electron 8035.T / TSE / Non-US | wafer fab equipment | 62.6 central #44 | 71,060 | $200.1B | 37 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #51 | ASML Holding ASML / NASDAQ ADR / Non-US | wafer fab equipment | 62.4 central #40 | 1,769 | $681.9B | 38 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | core AI winner, but already highly owned; score is capped by crowding and valuation discovery. | |
| #52 | Arm Holdings ARM / NASDAQ ADR / Non-US | custom ASIC / interconnect | 62.4 central #28 | 300 | $320.9B | 43 mapped 32 direct | In: Cash in from hyperscaler custom programs, data-center networking/optical conn... Out: Cash out to foundries, OSATs, IP/EDA, substrates, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #53 | Photronics PLAB / NASDAQ / US | materials / wafers | 62.0 central #22 | 28.14 | $1.79B | 61 mapped 20 direct | In: Cash in from recurring consumables and wafer/material volumes tied to fab uti... Out: Cash out to raw materials, energy, specialty chemicals, capex and process eng... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. | |
| #54 | Aixtron AIXA.DE / XETRA / Non-US | analog / power / RF | 61.9 central #76 | 44.62 | $7.80B | 20 mapped 20 direct | In: Cash in from component shipments to OEMs/distributors across industrial, auto... Out: Cash out to foundries/IDM fabs, OSATs, distributors, inventory and R&D. | selective AI/edge/power exposure; less direct than HBM, advanced packaging and AI networking names. | |
| #55 | KLA KLAC / NASDAQ / US | wafer fab equipment | 61.8 central #43 | 221 | $288.9B | 37 mapped 20 direct | In: Cash in from fab capex orders, tool shipments, spares, service contracts and ... Out: Cash out to components, precision subsystems, supply chain, R&D and manufactu... | levered to AI-driven fab and HBM capacity additions; upside depends on order timing and export-control risk. |
Every edge carries evidence, confidence, and cash-flow language.
This is the audit surface for the graph. The same visual connection can mean direct supply, capacity use, equipment sale, ecosystem adjacency, or competitive share shift, so the edge table keeps those distinctions visible.
Evidence mix
| Source | Target | Relationship | Cash flow | Evidence | Confidence |
|---|---|---|---|---|---|
AVGO Broadcom | MRVL Marvell Technology | competes in custom silicon/networking AI networking, custom silicon and optical interconnect | competitive share shifts among hyperscaler silicon suppliers not itemized | public product disclosures | high |
AVGO Broadcom | AMKR Amkor Technology | uses assembly/test ecosystem semiconductor assembly/test | Broadcom pays Amkor/OSAT providers; exact amount not disclosed undisclosed | Broadcom Form 10-K names Amkor | high |
AVGO Broadcom | NVDA NVIDIA | competes/coexists in AI data-center silicon custom AI accelerators, networking and GPU data-center architectures | hyperscaler capex split across GPU/custom ASIC/networking vendors not itemized | Broadcom and NVIDIA filings | high |
AVGO Broadcom | TSM Taiwan Semiconductor Manu... | uses foundry capacity custom AI accelerators/XPUs and networking silicon manufacturing | Broadcom pays TSMC for wafer fabrication; exact amount not disclosed undisclosed | Broadcom Form 10-K names TSMC as foundry supplier | high |
AVGO Broadcom | ASX ASE Technology Holding | uses assembly/test ecosystem OSAT assembly/test for semiconductors | Broadcom pays ASE/SPIL ecosystem providers; exact amount not disclosed undisclosed | Broadcom Form 10-K names ASE/SPIL as suppliers | high |
NVDA NVIDIA | 000660.KS SK hynix | buys memory HBM/DRAM in AI accelerator platforms | NVIDIA pays SK hynix or intermediaries for AI memory; HBM volumes/prices not itemized undisclosed | NVIDIA Form 10-K names SK hynix as memory supplier; SK hynix reports HBM AI demand | high |
NVDA NVIDIA | MU Micron Technology | buys memory HBM/DRAM in AI accelerator platforms | NVIDIA pays Micron or intermediaries for AI memory; volumes/prices not itemized undisclosed | NVIDIA Form 10-K names Micron as memory supplier; Micron reports HBM/data-center demand | high |
NVDA NVIDIA | 005930.KS Samsung Electronics | uses foundry and memory supply wafers and high-performance memory supply chain | NVIDIA pays Samsung ecosystem suppliers; exact allocations undisclosed undisclosed | NVIDIA Form 10-K names Samsung as foundry and memory supplier | high |
NVDA NVIDIA | AMD Advanced Micro Devices | competes in AI accelerators GPU/accelerator platforms for hyperscalers | competitive displacement risk; no direct cash flow assumed not a customer cash flow | NVIDIA and AMD public disclosures | high |
NVDA NVIDIA | TSM Taiwan Semiconductor Manu... | uses foundry capacity advanced GPU/accelerator wafer fabrication | NVIDIA pays TSMC for wafers; exact node/customer economics undisclosed undisclosed by vendor/customer | NVIDIA Form 10-K names TSMC as foundry supplier | high |
MU Micron Technology | 000660.KS SK hynix | competes in HBM/DRAM HBM, server DRAM and AI memory | competitive pricing and share shifts not customer cash flow | public memory-market disclosures | high |
MU Micron Technology | 005930.KS Samsung Electronics | competes in HBM/DRAM/NAND memory products for AI and servers | competitive pricing and share shifts not customer cash flow | public memory-market disclosures | high |
MU Micron Technology | NVDA NVIDIA | supplies memory ecosystem HBM/DRAM supporting NVIDIA AI accelerator platforms | NVIDIA/OEM ecosystem pays Micron for memory; direct amounts not itemized undisclosed | NVIDIA names Micron; Micron reports HBM/data-center records | high |
000660.KS SK hynix | MU Micron Technology | competes in HBM/DRAM AI memory pricing/share | competitive cash-flow shifts not customer cash flow | public memory-market disclosures | high |
000660.KS SK hynix | 005930.KS Samsung Electronics | competes in HBM/DRAM/NAND AI memory pricing/share and technology race | competitive cash-flow shifts not customer cash flow | public memory-market disclosures | high |
000660.KS SK hynix | NVDA NVIDIA | supplies memory ecosystem HBM/DRAM for AI accelerators | NVIDIA/OEM ecosystem pays SK hynix for memory; direct amounts not itemized undisclosed | NVIDIA names SK hynix; SK reports AI-led HBM demand | high |
005930.KS Samsung Electronics | 000660.KS SK hynix | competes in HBM/DRAM/NAND AI memory pricing/share and technology race | competitive cash-flow shifts not customer cash flow | public memory-market disclosures | high |
005930.KS Samsung Electronics | MU Micron Technology | competes in memory DRAM/NAND/HBM pricing and share | competitive cash-flow shifts not customer cash flow | public memory-market disclosures | high |
005930.KS Samsung Electronics | AMD Advanced Micro Devices | supplies/partners for HBM4/DRAM and programmable logic manufacturing HBM4 for MI455X and DRAM for EPYC; IC manufacturing | AMD/platform ecosystem pays Samsung; exact amount not itemized undisclosed | AMD Q1 FY26 results and AMD Form 10-K | high |
005930.KS Samsung Electronics | NVDA NVIDIA | supplies memory/foundry ecosystem HBM4/SOCAMM2 for NVIDIA Vera Rubin plus foundry/memory supply chain | NVIDIA/OEM ecosystem pays Samsung; exact amount not disclosed undisclosed | Samsung Q1 2026 results and NVIDIA Form 10-K | high |
AMD Advanced Micro Devices | GFS GlobalFoundries | uses foundry capacity larger-node microprocessor/GPU products and legacy nodes | AMD pays GlobalFoundries under wafer supply arrangements undisclosed | AMD Form 10-K names GlobalFoundries for larger than 7nm nodes | high |
AMD Advanced Micro Devices | 005930.KS Samsung Electronics | uses Samsung foundry/memory collaboration programmable logic ICs and HBM4/DRAM collaboration for AI platforms | AMD pays/partners with Samsung; exact split undisclosed undisclosed | AMD Form 10-K and Q1 FY26 results mention Samsung collaboration | high |
AMD Advanced Micro Devices | 2449.TW King Yuan Electronics | uses outsourced test services outsourced assembly/test services for semiconductors | AMD pays KYEC for ATMP/test services; exact amount not itemized undisclosed | AMD Form 10-K discloses KYEC as ATMP provider | high |
AMD Advanced Micro Devices | 002156.SZ Tongfu Microelectronics | uses ATMP joint-venture ecosystem outsourced assembly, test, mark and pack services | AMD pays Tongfu JV/related ATMP entities; exact amount not itemized undisclosed | AMD Form 10-K discloses Tongfu JV/ATMP dependence | high |
AMD Advanced Micro Devices | NVDA NVIDIA | competes in AI accelerators GPU/accelerator platforms for AI data centers | competitive revenue displacement; no direct cash flow assumed not customer cash flow | public products and filings | high |
AMD Advanced Micro Devices | TSM Taiwan Semiconductor Manu... | uses foundry capacity 7nm-and-smaller CPU/GPU/accelerator wafers | AMD pays TSMC for advanced-node wafers undisclosed | AMD Form 10-K names TSMC for all 7nm-and-smaller microprocessor/GPU wafers | high |
AMD Advanced Micro Devices | UMC United Microelectronics | uses foundry capacity programmable logic device ICs and specialty manufacturing | AMD pays UMC for IC manufacturing undisclosed | AMD Form 10-K names UMC | high |
INTC Intel | AMD Advanced Micro Devices | competes in CPU/data-center compute EPYC versus Xeon, AI server host CPUs | customer capex shifts between AMD and Intel not itemized | public product disclosures | high |
042700.KQ Hanmi Semiconductor | CRDO Credo Technology | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | ALAB Astera Labs | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | CAMT Camtek | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | 042700.KQ capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
042700.KQ Hanmi Semiconductor | NVMI Nova | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | 042700.KQ capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
042700.KQ Hanmi Semiconductor | MRVL Marvell Technology | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | 3661.TW Alchip Technologies | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | FORM FormFactor | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | 042700.KQ capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
042700.KQ Hanmi Semiconductor | ONTO Onto Innovation | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | 042700.KQ capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
042700.KQ Hanmi Semiconductor | 3443.TW Global Unichip | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | 6857.T Advantest | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | 042700.KQ capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
042700.KQ Hanmi Semiconductor | ARM Arm Holdings | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | AVGO Broadcom | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | AMD Advanced Micro Devices | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | 6526.T Socionext | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | NVDA NVIDIA | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | QCOM Qualcomm | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | 2454.TW MediaTek | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
042700.KQ Hanmi Semiconductor | TER Teradyne | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | 042700.KQ capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
042700.KQ Hanmi Semiconductor | COHU Cohu | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | 042700.KQ capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
042700.KQ Hanmi Semiconductor | AEHR Aehr Test Systems | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | 042700.KQ capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
CAMT Camtek | 000660.KS SK hynix | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
CAMT Camtek | MU Micron Technology | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
CAMT Camtek | GFS GlobalFoundries | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
CAMT Camtek | 005930.KS Samsung Electronics | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
CAMT Camtek | TSEM Tower Semiconductor | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
CAMT Camtek | TSM Taiwan Semiconductor Manu... | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
CAMT Camtek | UMC United Microelectronics | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
CAMT Camtek | 3105.TWO WIN Semiconductors | sells capacity-enabling tools/services fab/packaging equipment, service, yield, lithography, etch/deposition... | 3105.TWO pays Camtek for tools, spares, service or process technology; customer amounts generally undisclosed. not separately disclosed | mapped public-company AI semiconductor supply-chain relationship; cash-flow amount usually not disclosed unless separately cited | medium |
CAMT Camtek | 5347.TWO Vanguard International Se... | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
CAMT Camtek | 000990.KS DB HiTek | sells capacity-enabling tools/services fab/packaging equipment, service, yield, lithography, etch/deposition... | 000990.KS pays Camtek for tools, spares, service or process technology; customer amounts generally undisclosed. not separately disclosed | mapped public-company AI semiconductor supply-chain relationship; cash-flow amount usually not disclosed unless separately cited | medium |
CAMT Camtek | 2408.TW Nanya Technology | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
CAMT Camtek | 2344.TW Winbond Electronics | sells capacity-enabling tools/services fab/packaging equipment, service, yield, lithography, etch/deposition... | 2344.TW pays Camtek for tools, spares, service or process technology; customer amounts generally undisclosed. not separately disclosed | mapped public-company AI semiconductor supply-chain relationship; cash-flow amount usually not disclosed unless separately cited | medium |
CAMT Camtek | 6770.TW Powerchip Semiconductor M... | sells capacity-enabling tools/services fab/packaging equipment, service, yield, lithography, etch/deposition... | 6770.TW pays Camtek for tools, spares, service or process technology; customer amounts generally undisclosed. not separately disclosed | mapped public-company AI semiconductor supply-chain relationship; cash-flow amount usually not disclosed unless separately cited | medium |
CAMT Camtek | 2337.TW Macronix International | sells capacity-enabling tools/services fab/packaging equipment, service, yield, lithography, etch/deposition... | 2337.TW pays Camtek for tools, spares, service or process technology; customer amounts generally undisclosed. not separately disclosed | mapped public-company AI semiconductor supply-chain relationship; cash-flow amount usually not disclosed unless separately cited | medium |
CAMT Camtek | INTC Intel | sells into fabs capacity additions and process-node ramps in logic, memory and advanc... | fabs/IDMs pay CAMT for equipment/materials/services mostly undisclosed | company reports + equipment supply chain | medium |
AMKR Amkor Technology | CRDO Credo Technology | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
AMKR Amkor Technology | ALAB Astera Labs | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
AMKR Amkor Technology | CAMT Camtek | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | AMKR capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
AMKR Amkor Technology | NVMI Nova | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | AMKR capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
AMKR Amkor Technology | MRVL Marvell Technology | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
AMKR Amkor Technology | 3661.TW Alchip Technologies | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
AMKR Amkor Technology | FORM FormFactor | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | AMKR capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
AMKR Amkor Technology | ONTO Onto Innovation | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | AMKR capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
AMKR Amkor Technology | 3443.TW Global Unichip | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
AMKR Amkor Technology | 6857.T Advantest | uses inspection/test tools advanced package inspection, handlers, probe and reliability test | AMKR capex flows to test/metrology/tool vendors undisclosed | equipment supply chain | medium |
AMKR Amkor Technology | ARM Arm Holdings | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
AMKR Amkor Technology | AVGO Broadcom | serves AI/fabless customers assembly, test, advanced packaging, chiplet integration | AI chip companies pay packaging/test providers mostly undisclosed | OSAT supply-chain map | medium |
Depth-five paths show second-order exposure without pretending it is first-order revenue.
The path table is useful for finding shared dependencies: a node may not sell directly to an AI accelerator company, but it may connect through foundry capacity, OSAT, metrology, equipment, memory, or materials.
Most path-rich sources
The named-focus list is a compact diligence queue.
These seven rows preserve explicit verdicts, key relationships, risks, revenue streams, and cash summaries from the bundle.
SK hynix
Best named-seven alpha: direct HBM scarcity and record profitability; crowding rising after record share move.
Micron Technology
Strong memory-cycle alpha: HBM/data-center mix, margin expansion and contracted supply make it a top named-seven candidate.
Samsung Electronics
Improving AI linkage: HBM4/SOCAMM2 and foundry wins matter, but conglomerate scale dilutes pure-play alpha.
Broadcom
High-quality but less early: custom AI silicon and networking cash flow are real, valuation and concentration cap alpha.
Advanced Micro Devices
Selective alpha: better entry leverage than NVIDIA if Instinct/EPYC ramps convert, but execution risk is high.
NVIDIA
Hold/compounder more than fresh alpha: extraordinary AI earnings power, but the market has already discovered it.
Intel
Turnaround option: foundry/packaging upside, but cash-flow evidence is weak versus losses and execution risk.
The source ledger stays attached to the analysis.
The bundle combines filings, investor relations pages, company results, ETF holdings context, and relationship-map evidence. The links below preserve source trails and raw data for audit.
Cited source ledger
Download the research tables
Source interpretation note
Relationship rows are not automatically revenue rows. The cash-flow field distinguishes direct customer economics, supplier outflows, competitive shifts, and undisclosed amounts. The confidence field should be read before converting an edge into a valuation input.